boost posix time question

Hello

I'm developing server and client applications and I need ping-pong mechanism to get ping time betwen them (time needed for data to arrive)..

I've been thinking the best way would be to send

Code:

---

boost::posix_time::microsec_clock::universal_time()

---

from server to client, which would then just reply with its own universal_time() ..


The first problem is I dont know how to store universal_time() as an int/long.. I know I can store it as a string but this would take extra cpu to parse the data each time (with boost posix library)..

All the data I send is serialized - maybe I could just serialize the

Code:

---

boost::posix_time::ptime

---

object and send over?


The next thing is what if server's time zone is gmt+1 and client's gmt+3? Will that affect the the time if I do the subtraction to get the ping time? If so, what can I do about it?

Maybe anyone has better solution in mind?

Thanks a lot for help

universal_time() is independent of timezone. It's not independent of inaccurate clocks, though.

To measure travel time of packets, a peer can only rely on itself. What you do is, you send a very simple packet (ping uses the ICMP protocol and the ICMP_ECHO_REQUEST message) to the server, which responds to it by immediately sending an also very simple response packet (ICMP_ECHO_RESPONSE in ping). Turnaround time should be as close to zero as possible. The client marks the time when it sends, and when it receives the answer. It then has the full time it takes for the request-response cycle. It can repeat it a few times to get an average. It can divide it in half to get the one-way time - on average, that will be accurate.
If the server needs the time, too, it can either do its own ping or, more efficiently, you can make a pong protocol, i.e. clients sends ECHO(1), server responds with ECHO(2), client responds immediately with ECHO(3), and so on, up to, say, ECHO(7). The client measures between send(1) and rcv(2), send(3) and rcv(4), send(5) and rcv(6), while the server measures between send(2) and rcv(3), send(4) and rcv(5), send(6) and rcv(7).

Or something like that. But never rely on another computer's clock, unless they are in a time-synchronized environment.

Quote:

---

unless they are in a time-synchronized environment

---

I do not know what exsactly you mean by this... by even if 2 computers have "synchronized" their times using some time-synchronization mechanism - difference can still be same order as a trip time between two computers or between them and time-server...

So I will not relay on these two times for packet trip mesurement... Round-Trip approach is OK

Time protocols are not simple, but they do manage to minimize the inaccuracy that results from the lag.

Thanks a lot!